Learning Experiences with Chat GPT

Posted on 1 Comment on Learning Experiences with Chat GPT

The new, AI-powered search engine Chat GPT has been the subject of many news stories since it launched a few months ago. To learn more about AI-powered search and how it has the potential to transform teaching, learning, and research in higher education, I attended what may have been the world’s first training workshop on the tool. It was presented by Phil Bradley, a librarian and consultant who writes and teaches about internet searching and social media in libraries.

As someone whose professional moniker is ‘The Internet Librarian’, Bradley was infectiously excited about the transformative potential of this new technology, while emphasising its risks and limitations. Here are seven things I learned from his excellent workshop!

  1. The GPT in Chat GPT stands for Generative Pre-trained Transformer. It is a chatbot, funded by the company OpenAI, and launched in November 2022. It uses a large language model (LLM), GPT-3, to respond to user queries or requests. Training it involved an enormous quantity of webpages which date from pre-2021; OpenAI employees then asked it a series of questions and scored its answers. Chat GPT then used this scoring to learn and improve its responses.
  2. Currently, when we use Google or another search engine for a question like, ‘What is a reliable journal in medieval history?’ we get a list of webpages, and need to analyse, sort, and synthesize the information presented for ourselves. Chat GPT looks nothing like this—it will provide a list of journals and information about them and it will retain and remember the query, so we can build on initial results for further searching. For example, a user might ask Chat GPT to put the initial list of journals it provided into a table and include URLs (this can be done in a single request) or might ask that the results focus on Open Access journals.
  3. Chat GPT is a particularly good tool for users who struggle to identify the terms they should use to find the information they need—a user can ask it to suggest search terms to use, or journals or databases to search.
  4. Chat GPT does not, currently, provide trustworthy content. The LLM used to train it has trained it to be verbose—when it does not know the answer it will make something up! A user can ask it a factual question and it will reply with places or publications that do not and have never existed. This, of course, is particularly concerning when it is used by people who do not know that they should, or do not want to, verify and cross-check the information it provides.
  5. It will generate new content on request—Phil Bradley’s brother, an HR professional, asked Chat GPT to write a letter rejecting someone for a job, and thought it was an excellent letter. A key thing to note is that this content is not plagiarized or taken from somewhere on the web—the text is generated by the AI, and will be different every time the request is repeated. This raises challenges for citing the results it gives.
  6. The field is developing incredibly rapidly, and investors are pouring an enormous amount of money into it. Microsoft is planning to launch a Chat GPT integration in March, and there are plans afoot to incorporate it into the search engine Bing. It took five days to reach a million users and due to high traffic can lag and be quite slow ( UK-based users are advised to play with it in the morning, before people who live in North America get online).
  7. You can try Chat GPT, and other similar tools, at the following links. Note that you need to create an account to use Chat GPT; this is currently free but a paid version is coming, and trajectory for the future seems to be that companies will have both a free and a subscription version of their product.

The workshop chat was very lively and many people dropped links to recent articles they had found helpful or interesting. Here are some of these.

I’m hoping to explore Chat GPT further, from the twin perspectives of a humanities librarian and medieval studies researcher, over the coming weeks. In the meantime, if you are currently trying out AI search engines, I’d be curious to hear how you think they might impact your teaching, learning, or research.

One thought on “Learning Experiences with Chat GPT”

Leave a Reply

Your email address will not be published. Required fields are marked *